Getting started with Claude
Where to use it, how to start, and what to keep in mind — for everyone, not just developers.
Claude is an AI assistant. You can chat with it like you would with a colleague, or you can let Claude Code work directly in your files and projects — reading, drafting, and editing for you. This page shows the two ways to work with it, where you can open it, how to get going in a few minutes, and a few habits that keep it safe and useful at Qvantum.
Two ways to work with Claude
Most confusion comes from mixing these up. Pick the one that matches your task — you can use both, and the same Qvantum account works across them.
Claude — chat
Ask questions, write and rewrite text, summarise documents, brainstorm, translate, or analyse a spreadsheet or image. Nothing to install for the web version — open a browser and start typing.
Reach for it when: you want an answer, a draft, or a second opinion.
Claude Code — the agent
Claude works inside a folder on your computer: it reads your files, proposes changes, runs commands, and edits documents or code for you — one approved step at a time. Far more capable, and the focus of the rest of this page.
Reach for it when: the work lives in files — code, sites, documents, data.
Where you can open it
Claude meets you where you already work. The chat is the same everywhere; Claude Code runs in the terminal, in your editor, or in the browser.
claude, and start working. The most powerful way to use it.Get started in four steps
From nothing to your first task in a few minutes. Steps 2–4 are for Claude Code in the terminal — for a full walkthrough, see Install Claude Code.
Chat for questions and drafts; Claude Code when the work is in files.
Install Claude Code, then sign in with your Qvantum Claude account (your browser opens once — after that it remembers you). Don't have a license yet? Get a license first.
Go to the folder you want to work in, start Claude, and let it learn the project once with
/init — that writes a CLAUDE.md file it remembers next time.
Describe the task in plain language. Review what it proposes and approve — or turn on “auto mode” (below) so it keeps moving without asking for every change.
Turn on “auto mode”
By default Claude Code asks before every change — safe, but slow once you trust it. One shortcut lets it apply its edits without stopping each time. This is what people mean by “auto mode on”.
The shortcut
Inside Claude Code, press Shift+Tab to cycle the permission mode. Stop on “accept edits on”.
Each press of Shift+Tab moves to the next mode:
A fourth press cycles back to normal. The little ⏵⏵ accept edits on line at the bottom of the input box is your sign that auto mode is active (exact wording can vary slightly between versions).
Prefer to start in auto mode? Launch Claude Code with this flag — it begins with edits auto-accepted:
When to use which
- Auto mode (accept edits on) — best when you're iterating and watching the screen. Claude keeps moving; you skim the changes as they land.
- Plan mode — best for anything big or unfamiliar. Claude explores and proposes a plan before touching a file. Read it, then approve. Start there with
claude --permission-mode plan. - Normal — best for sensitive or one-off work where you want to confirm every step.
One caution
There is also a fully hands-off option,
--dangerously-skip-permissions, that skips all prompts. Keep it for throwaway, isolated
environments only — never on sensitive Qvantum work. Auto-accept edits is the everyday choice.
What to keep in mind
Data first
Using Claude is safe — your data is kept safe, is
not used to train AI models, and should not leave the EU. Even so, treat it
like any tool: don't paste passwords or API keys (keep those in a .env file, never in chat), and
check with your manager or IT before sharing customer personal data or patent-sensitive material.
- Give it context. Run
/initonce per project to create aCLAUDE.md— Claude's memory of how your project works. - Plan before big changes. For anything large, use plan mode first. Reading the plan takes a minute and saves ten.
- You are the gate. Claude drafts and proposes; you approve. Skim the changes even in auto mode — nothing ships without you.
- One task at a time. Use
/clearbetween unrelated jobs so old context doesn't muddy a fresh task. - Be specific. Say what “done” looks like. Clear goals and examples beat clever phrasing every time.
- Match the mode to the risk. Auto-accept while iterating; normal mode for sensitive or destructive work. Switch any time with Shift+Tab.
Next steps
Need a seat first? Get a license. Want the full terminal setup? Read Install Claude Code. Stuck on something? Ask an ambassador.
Maintained by Dominic Sandner · Last updated 2026-06-24